Agent Claude Code
FIND lens (Round 2) — refutes the claim that every promise in the abstract/summary is delivered, in matching strength, by the body.
Agent Claude Code
FIND lens (Round 2) — refutes the claim that every promise in the abstract/summary is delivered, in matching strength, by the body.
Agent Claude Code
FIND lens (Round 5) — refutes the claim that the prose is free of AI-generated verbal tics that signal unedited machine writing.
Agent Claude Code
FIND lens (Round 6 / release gate) — refutes the claim that every quotation is verbatim and every citation says what the source actually says. Re-fetches primary sources, full text.
Agent Claude Code
FIND lens (Round 2) — refutes the claim that each causal/quantitative/scope assertion is backed by the specific evidence it points to.
Agent Claude Code
FIND lens (Round 6, DELTA-audit) — refutes the claim that the Round 2 calibration fixes held AND that the rewriting introduced no new overclaim.
Agent Claude Code
FIND lens (Round 2 / Round 3) — refutes the claim that every internal reference resolves and every bib entry is real. Owns orphans and phantoms.
Agent Claude Code
FIND lens (run MID-ladder AND FINAL / code profile) — generalist adversarial read-through that owns "technically wrong description of the domain object.".
Agent Claude Code
FIND lens (Round 4) — refutes the claim that figures, their captions, the numbers inside them, and the prose that describes them all agree.
Agent Claude Code
FIND lens (Round 1 / release gate) — refutes the claim that the artifact leaks no proprietary value, secret, or restricted detail across the stated boundary.
Agent Claude Code
FIND lens (Round 5 / protocol profile) — refutes the claim that the limitations section is complete. Its specialty is what is missing, not what is present.
Agent Claude Code
FIND lens (Round 1 / release gate) — refutes every number in the artifact by RECOMPUTING it from canonical data. Reading a summary is never verification.
Agent Claude Code
FIND lens (Round 3) — refutes the claim that a multi-artifact series tells one non-contradictory story when read in sequence.
Agent Claude Code
FIND lens (Round 5) — refutes the claim that the prose is clear, well-structured, and free of redundancy, dead transitions, and jargon fog.
Agent Claude Code
FIND lens (protocol / release gate) — refutes the claim that the artifact's provenance and pre-registration story matches the repository's own history.
Agent Claude Code
FIND lens (Round 6) — refutes the claim that the artifact's positioning is complete and its novelty is not already scooped by recent, dated literature.
Agent Claude Code
FIND lens (Round 4 / release gate / code profile) — refutes the claim that a reader could reproduce the results from what the artifact provides.
Agent Claude Code
FIND lens (paper / protocol profile) — refutes the claim that the statistical methodology is appropriate and its reported statistics are plausible on recompute.
Agent Claude Code
FIND lens (Round 7 / release gate) — refutes the claim that the artifact is mechanically ready to ship. Every item is PASS or FAIL, decided by execution.
Agent Claude Code
FIND lens (Round 3) — refutes the claim that every term, symbol, and unit means one thing consistently within and across the artifacts.
Agent Claude Code
VERIFY stage (all profiles) — takes ONE finding plus artifact paths and reproduces it or demotes it. CONFIRMED requires recompute/refetch/execution evidence; default PLAUSIBLE when uncertain.
Skill Claude CodeCodex
Run The Adversary v2 — the three-stage adversarial audit (CHECK deterministic, FIND independent lenses, VERIFY reproduce-or-demote) over an artifact at a pinned commit. Use when the user says "run the adversary", "adversarial audit", "audit this paper/repo", "release gate", or "gate this before I ship". Picks the…
Instructions file
Instructions for tasumermaf/the-adversary, covering the adversary — repo identity for claude code, your role as the hub, the rules in force, the engine (never reorder) and where things live.